Data AI Weekend Sessions
more to be announced
Between the Strategy Deck and the Invoice: Data Platform and AI Economics for 2027
Kierra Dotson
Karl Ivo Sokolov
Most AI and data platform decisions are made in one language and paid for in another. Strategy gets written in decks and roadmaps, and the bill arrives as run-rate, licences, inference cost and headcount that nobody modelled.
Kierra and Karl Ivo come at that gap from opposite ends. Kierra opens upstream of the business case, where the economics are largely set before anyone calculates ROI: what you are actually building, which capabilities you end up owning, and how much of the case rests on conditions the company does not control, from model and cloud dependencies to where data is allowed to move.
Karl Ivo comes in from the CFO side, where that investment meets financial machinery built for deterministic projects. No slides, no exercises, no breakout groups, and everyone gets a turn to talk. Come with one decision you are currently making or deferring, and one number you expect Finance to ask you for.
90 minutes. Tentative format: three discussion rounds of roughly twenty minutes, moving from what this really costs, to what value is actually being bought, to what should change in the 2027 plan.
What does "creation" mean in the age of AI?
I recently used AI in a workflow that helped me write five novels in twelve months. I didn't have AI write the words in the novels, only act as my editorial and research staff. But I'm less interested in prescribing my process than in comparing notes. How are other people using AI in creative work whether authorial or technical? What helps? What new problems does it create? At what point does human-in-the-loop stop being enough to qualify it as "creative" or "yours"? Does that threshold change when we're asking what is morally acceptable versus what is acceptable as a product?
What is teaching and what is learning in the age of AI?
I've been working on learning paths meant to take a learner from zero to hero in a skill. Creating those paths is heavily assisted by AI, and theoretically someone can use AI to produce strong responses to prompts and quizzes along the way. This makes me think of a professor I knew who received quiz answers that were almost all clearly AI-generated. The wrinkle was that he had created the quiz with AI. What does it mean to say a teacher has taught a skill or a learner has learned it when agents are increasingly interacting with other agents rather than end users?
If AI stops us breaking things, where does judgement come from?
Almost everything I know as a self-taught engineer, I learned by breaking production, making mistakes. Bring your worst pre-AI failure. Let's figure out which lessons we learned the hard way. How do we share them with the next generation, especially if they never face those challenges or just hit "next" into the prompt?
If AI eats the junior work, how does anyone get in the long way round?
I've been talking to folks across engineering, data, and more in a unique way. They began with a spreadsheet that no one wanted then, this became their first script or integration. They created an internal tool no one asked for, but it saved hours of work each week. That's exactly the work AI is absorbing. I want to figure out where the side doors are now, and whether we need to build new ones.
How can a company measure whether AI is being used well?
A friend of mine, not in tech, told me her company was pushing everyone to use AI. She was frustrated. She said: "I am not against AI. It has very cool features, but I don't care about cool. I care about getting my job done. I went back and forth using AI for something at work, it consumed my time, and I ended up doing my work manually because it was faster."
Everyone is being told to use AI. Almost no one is being told how to tell whether it's working.
Tokens cost money. Time costs more. And "we're an AI company now" is not a metric. Somewhere between the mandate from above and the person quietly doing the work by hand because it's faster, there's a question nobody has answered: what does using AI well actually look like, and how would you know?
Where does AI actually belong in the workflow?
I keep seeing companies start with the AI tool and then try to figure out where people can use it. I'd love to flip that around and have a conversation about starting with the actual work instead. Where are people losing time? Where are they getting stuck? What parts could AI make easier, and what parts still need human knowledge and judgment? Basically: map the workflow first, find the friction, and then ask where AI belongs.
This follows my earlier session, How can a company measure whether AI is being used well?
Will the real semantic layer please standup
Hot take, the only data migration that makes sense is to Apache Iceberg, but what comes next after you've chosen your storage layer?
What does Openness in Open Source mean nowadays?
There tends to be a lot of negativity between project A and project B that closes off communities, even though it is easier than ever to simply work together. This is many times the marketing machines at play, but I want to get people's opinions as well. It is so much easier than ever to work together, and if AI could do one thing, it could be used to bridge the gap between projects, vs isolate and disrupt.
The Future of Distributed Data Systems
I believe we're on the path to a thesis proposed by James Urquhart in his book Flow Architectures that was a more fluidic internet, and I feel like we are getting there with the advances in agentic infrastructure. I'd love to see what other people believe when it comes to the next 1 to 3 years.
Does your reporting data layer live in 5 different tools?
How can we handle unified insights in an era where analytics are more accessible and an organization's analytics layer can be spread between many tools in the tech stack?
AI as Mathematician: how it works, when it doesn't, and what it means for other disciplines
Culture and technology, a dynamical system: what the hell is even happening anymore?
Agents of Chaos
Are agents trustworthy delegates for business tasks, relentless hacking machines or something else entirely? Let's discuss the competing narratives around agentic technology. Let's also discuss the tools we use for directing agentic behavior, such as context and harnesses.
Getting Business Value From AI : pt 2 / AMA
How this session came about: A Weekend with Bill Inmon
Getting Business Value From AI : A Workshop
About the workshop: A Weekend with Bill Inmon
Who should own an AI agent's memory?
As agents move from one-off chat sessions to long-running work, memory becomes more than convenience. It becomes a governance problem: what should be remembered, who can inspect it, who can correct it, when it should expire, and whether it should live inside a vendor tool, a team-owned context layer, or the user's own portable memory. Furthermore, as organizations deploy more agents, each one building its own private memory feels brittle and risky. In this session, I want to explore whether memory should become shared infrastructure, with policies for provenance, access control, evaluation, correction, and portability across models and agentic harnesses.
Can smaller models make agents more trustworthy?
For this session, I'm advocating the reliability benefits of using smaller, specialized models for bounded decisions inside agentic systems. Reasoning traces can look persuasive while remaining unstable, incomplete, or hard to evaluate. I want to explore whether trust should come less from reading a model's chain of thought and more from surrounding the model with smaller, inspectable decision points that constrain where the agent can go next. If routing, guardrails, scoring, and tool selection become explicit classification problems, we may get systems that are easier to test, monitor, debug, and improve.
I'd also connect this to the Jev hype. Tools like Jev point to a useful pattern: fast, structured classification inside the agent loop. But I'm not convinced every team needs a general-purpose classification model when many routing, scoring, and guardrail tasks are narrow enough to train or tune a smaller classifier with local data.
AI makes us wrong faster. How does that help the business? Who succeeds the most in this environment?
How does data infrastructure need to evolve when agents drive the queries?
Through the Context Glass: A Deep Dive on Principles
Jessica Freasier
Everyone is talking about context, but what actually is it? In the session, we will be stepping outside the context of "context in AI" and taking a look more generally at definitions, roles, and characteristics in context (regardless of whether this is fed to an LLM).